Why there is no single detector
JCQ is a membership body of the UK’s leading qualification providers — AQA, OCR, Pearson, WJEC and CCEA among them — and it writes the rules those exam boards apply to schools. Its AI rules sit in one document: AI Use in Assessments: Your role in protecting the integrity of qualifications. First published on 26 April 2023, it was revised on 2 February 2024 and again on 30 April 2025, and that second revision is still the version JCQ’s AI page links as current. You will also see it cited as “AI Use in Assessments: Protecting the Integrity of Qualifications”, which is how Ofqual and JCQ’s own later documents shorten it.
The cover says who it is written for: “Heads of Centre, Teachers and Assessors”. Not students, and not software. Its scope is equally specific. Because candidates sitting an exam have no internet access, JCQ explains that the document is about the work done outside the exam hall:
“The majority of these assessments will be Non-Examined Assessments (NEAs), coursework and internal assessments for General Qualifications (GQs) and Vocational and Technical Qualifications (VTQs).”
The obligation it places is on people. Teachers and assessors “must only accept work for qualification assessments which is the students’ own”, and where they doubt that, “they must investigate and take appropriate action.” Nothing in it tells a school which product to buy, or whether to buy one.
JCQ’s instructions for conducting non-examination assessments, effective from 1 September 2026, turn that into signatures. They apply to “GCE AS, A level and GCSE qualifications accredited in England or Wales” with an NEA component. Every candidate signs a declaration that the work is their “own unaided work”; the teacher signs a separate authentication once it is finished. If the teacher cannot confirm the work is the candidate’s own, the instruction is blunt: do not accept it, and record a mark of zero for internally assessed work.
This page covers England, Wales and Northern Ireland. Scotland’s qualifications are run by SQA under its own rules, which we have not reviewed here. One gap for Northern Ireland: the NEA instructions quoted on this page do not mention CCEA, and the separate coursework instructions name only CCEA’s unitised AS and A levels. We have not confirmed which JCQ document governs internal assessment in CCEA GCSEs, so if you sit those, check with CCEA and your teacher.
The four detection tools JCQ names
Section 8 of the guidance (“Identifying misuse in assessments”), under the subheading “Automated detection”, describes programs that statistically analyse written content and “determine the likelihood that it was produced by AI, for example”, and then lists four:
| Tool, as JCQ lists it | Our explainer on how it works |
|---|---|
| Copyleaks | Copyleaks AI detector |
| GPTZero | GPTZero |
| Sapling | Sapling AI detector |
| Turnitin AI writing detection | Turnitin’s AI writing indicator |
“For example” is doing real work in that sentence. A few paragraphs later JCQ removes any doubt:
“The list of certain suppliers of AI-related products is for information purposes only and does not constitute an endorsement by JCQ. It is each centre’s responsibility to investigate and verify any suppliers they use … The list provided is not exhaustive.”
JCQ also attaches its own warnings to the tools, and they are worth reading closely:
- Edited AI text scores lower. The tools “will give lower scores for AI-generated content which has been subsequently amended by students, as they base their scores on the predictability of words.”
- Accuracy varies. It depends on “the AI tool and version used, the proportion of AI to human content, prompt types and other factors (such as an individual’s English language competency).” That last factor is the pattern behind non-native English writers being flagged more often.
- One tool is not enough. Where misuse is suspected “it may be helpful to use more than one detection tool to provide an additional source of evidence”.
- A score is one input. Detection “should form part of a holistic approach”, “all available information must be considered”, and teachers “are best placed to assess the authenticity of work submitted to them”.
JCQ’s newer summary for teachers (uploaded in September 2026; its file name says it is effective from September 2026, though the PDF itself carries no date) keeps the tools optional. Its wording is “Consider the use of AI detection tools and discussing the work with the student as part of a holistic approach” — consider, not run. Ofqual’s advice note to awarding organisations (27 April 2026, England) frames “statistical or technological tools” as a way to “help identify potential cases for investigation”, and warns that moderators “overly focused on identifying AI-related malpractice” have less attention left for the rest of moderation.
So a school’s own checks are not the only line of checking. Moderators and examiners see submitted work too, and two of the cases JCQ publishes, set out under the sanctions below, surfaced outside the classroom: one when a moderator raised concerns, one in a review for AI misuse that ran before the centre was contacted.
What a teacher looks for besides a score
The same section of the guidance lists potential indicators of misuse, lettered a) to p), and tells teachers to compare new work with what the student has produced before. Four of them, quoted:
- “A default use of American spelling, currency, terms and other localisations.”
- “Inclusion of references which cannot be found or verified (some AI tools have provided false references to books or articles by real authors).”
- “The inadvertent inclusion by students of warnings or provisos produced by AI to highlight the limits of its ability, or the hypothetical nature of its output.”
- “The submission of student work in a typed format, where their normal output is handwritten.”
None of those needs software. They are the reason a teacher who has watched a project grow over months sits at the centre of JCQ’s model, and a detector sits at the edge. Teachers wanting the wider picture can read our guide to AI detectors for secondary-school teachers, though it is written around US schools and law, or our general guide for teachers. Neither replaces JCQ’s rules for a JCQ centre.
What JCQ counts as AI misuse
JCQ defines the offence before it lists examples. AI misuse is “where a student has used one or more AI tools but has not appropriately acknowledged this use and has submitted work for assessment when it is not their own.” The examples, in the guidance’s own words:
- “Copying or paraphrasing sections of AI-generated content so that the work submitted for assessment is no longer the student’s own.”
- “Copying or paraphrasing whole responses of AI-generated content.”
- “Using AI to complete parts of the assessment so that the work does not reflect the student’s own work, analysis, evaluation or calculations.”
- “Failing to acknowledge use of AI tools when they have been used as a source of information.”
- “Incomplete or poor acknowledgement of AI tools.”
- “Submitting work with intentionally incomplete or misleading references or bibliographies.”
The list ends with the sentence that gives it force: “AI misuse constitutes malpractice as defined in the JCQ Suspected Malpractice: Policies and Procedures”. That policy, effective from September 2026, names both routes by which AI trouble usually arrives: plagiarism, including “incomplete referencing (including the internet and artificial intelligence (AI) tools)”, and “making a false declaration of authenticity” about coursework or NEA.
Note how often the word paraphrasing appears in that list. Running AI output through a rewording step does not take it off the list; it is the first item on it.
The NEA instructions add one more rule. Work must be in English or Welsh, as the awarding body specifies, and “AI translation tools must not be used to generate work for assessment.”
Before and after you sign the declaration
For a student, timing decides who handles a concern. The declaration you sign is the line between a problem your school deals with and a case the exam board hears.
| When the concern arises | Who deals with it | What JCQ says |
|---|---|---|
| Before you sign the authentication declaration | Your school or college, under its own malpractice policy | “There is no requirement to report the irregularity to the awarding body” — unless it breaches the conditions of the assessment |
| After you have signed, if your school raises it | The awarding body | “The head of centre must notify the relevant awarding body immediately using Form JCQ/M1”; and, as a separate point, “If malpractice is found, the awarding body will apply a sanction” |
| After you have signed, if an examiner or moderator raises it | The awarding body, which can involve your school | “As part of the malpractice investigation, the awarding body may ask the head of centre to gather information and report their findings.” |
That split appears in the NEA instructions (section 9), in the AI guidance itself, and in the teacher summary, which puts it in two lines: if the student hasn’t signed, “follow your school or college’s malpractice policy”; if the form is signed, “report to the awarding body”. CCEA tells Northern Irish students the same thing in plainer words.
The 2026/27 candidate declaration carries an AI clause of its own. You declare the work “has clearly acknowledged any use of AI”, alongside the instruction “You must clearly acknowledge and reference any use of AI”, and you sign that “if I make a false declaration this is a form of malpractice.” JCQ’s information sheet for candidates spells out the practical consequence: “If you have used an AI tool, you must declare that you have used it before signing the declaration form”.
So if you have used AI and not said so, the moment to say so is before your signature goes on that form, not after a teacher raises it.
The sanctions
The NEA instructions list four sanctions the awarding body may impose for a breach found after signing:
- the piece of work is awarded zero marks;
- disqualification from that component for that examination series;
- disqualification from the whole subject for that series;
- disqualification from all subjects, and being barred from re-entering for a period of time.
The malpractice policy is more granular. Its Appendix 6 sets nine standard sanctions against candidates: a warning; loss of marks for a section, a component or a unit; disqualification from the unit, from all units in one or more qualifications that series, from the whole qualification, or from every qualification taken that series; and a bar on entering for a set period. In its indicative tariff, plagiarised material that “comprises the substance of the work” sits in the band from sanction 5 (disqualification from the unit) to sanction 9.
These are not hypothetical. Appendix A of the AI guidance publishes real cases, chosen, JCQ says, to include both concerns centres reported and issues “awarding body assessment personnel have identified”. In an A Level History NEA with AQA as the awarding body, a teacher’s doubts led the centre to run AI detection software; one candidate “was disqualified from the A Level History qualification” and another received “a loss of all marks gained for the A Level History NEA component.” In an OCR Cambridge Nationals unit, a moderator raised concerns about missing referencing, and “through using Turnitin” two candidates were identified; both got zero marks for the affected learning objectives. And a Pearson Extended Project was “identified by detection software” during “a regular review of work for the purposes of identifying potential AI misuse”, before the centre was contacted. That candidate was disqualified from the qualification.
The published numbers are small but climbing, and they cover England only. Ofqual’s summer 2025 malpractice statistics record that plagiarism involving AI “accounted for 75.0% of all student plagiarism cases (100 cases) and 2.0% of all student malpractice cases”, up from 55.4% (85 cases) and 1.7% in summer 2024. Figures for summer 2026 are provisionally due on 10 December 2026.
Your rights if you are investigated
The malpractice policy gives candidates the right to be made aware of “all evidence that has been obtained during the investigation which supports the allegation”, and to submit a written statement in response. “Appeals must normally be made within 14 days of receiving the malpractice outcome decision.” If you wrote the work, your drafts, notes and version history are the evidence that matters, and how to show you wrote something yourself walks through gathering them.
Using AI and saying so
JCQ does not ban AI outright; it bans unacknowledged AI and AI doing the work that earns the marks. Section 4.3 of the 2026/27 NEA instructions sets the referencing rule:
“Where AI has been used, the reference must show the name of the AI tool used and should show the date the content was generated. For example: ChatGPT 3.5 (https://openai.com/blog/chatgpt/), 25/01/2027.”
That is only the first requirement. In the same section, candidates must also:
- acknowledge the sources the AI tool itself used in generating the content;
- keep the questions, the prompts and the AI output “in a non-editable format (such as a screenshot)”;
- provide “a brief explanation of how it has been used”, submitted with the work for final assessment.
The AI guidance closes two loopholes. Writing just ‘AI’ or ‘ChatGPT’ is “unacceptable… just as it would be unacceptable to state ‘Google’”. And acknowledging AI does not buy marks for it: the assessor “must still ensure the student is not rewarded if they have used AI tools such that they have not independently met the marking criteria.” JCQ’s student sheet says it in capitals: “IF YOU USE AN AI TOOL, YOU CANNOT GET MARKS FOR WHAT THE AI TOOL HAS PRODUCED”.
The EPQ, the Entry Level Certificate and CCEA’s unitised AS and A levels run under JCQ’s separate instructions for conducting coursework, also effective 1 September 2026, and they carry the same wording on tool name and date.
A couple of pages on this site can help with the paperwork, though neither replaces the screenshots JCQ asks you to keep. Our AI disclosure statement generator drafts a plain-language account of what you did, which is a starting point for the “brief explanation”. And how to cite ChatGPT covers the reference formats universities use. For JCQ work, follow JCQ’s format above, and your teacher’s instructions over both.
What the exam boards add
The boards work inside the JCQ rules rather than beside them. What each has published for students and teachers, as we found it on 26 September 2026:
| Board | Page and date | What it adds |
|---|---|---|
| AQA | Blog post, 15 February 2024 | Written for JCQ’s February 2024 revision, it points centres to the JCQ guidance and notes that revision added “an extended list of AI tools and AI detection tools”. We found no separate AQA rulebook for AI when we searched on 26 September 2026. |
| OCR | Subject update, first published 31 March 2026, updated 29 June 2026 | Adding an “AI use in the NEA” section to specifications and an “AI use in the EPQ” section to its Extended Project guides, and reminding teachers to supervise enough to authenticate the work. |
| Pearson (EPQ) | Teacher guide, Version 1.0, October 2023 | “The sole use of AI tools cannot meet assessment descriptor requirements”; every instance of AI use must be referenced, and failure to do so “will be considered malpractice.” |
| Pearson Edexcel (A level History) | Coursework guidance, undated | Students “should not be using AI tools” to select works, to plan, or to help write the assignment; the authentication sheet now refers to AI tools. |
| WJEC / Eduqas | Student guidance, undated | Suspected misuse triggers a malpractice investigation; students should “include screenshots of the generated content with a date and time stamp”. |
| CCEA | Student page, undated | Warns that misusing AI “even accidentally” could mean lost marks or disqualification from the subject. |
The pattern is consistent. No board page we read names a detector or requires schools to run one, although the Pearson and OCR cases above show boards checking work themselves; the variation is in how tightly each subject restricts AI, with Pearson’s History guidance the strictest we found. Check your own specification and your subject teacher’s instructions, because they can be narrower than JCQ’s baseline.
Where Ofqual stands
Ofqual regulates qualifications in England, and every Ofqual page cited here covers England only. JCQ’s malpractice policy names the other two regulators: Qualifications Wales, and CCEA Regulation for Northern Ireland. Ofqual’s statement on regulating AI, updated 16 July 2026, says undisclosed AI in coursework “damages the integrity of the qualification, as marks may no longer fully reflect the student’s own work”. It acknowledges that “the number of detected cases remains relatively low”, and lists among the steps it has taken writing to exam boards “to require stronger deterrence, detection and authentication measures”. The Ofqual guide for schools and colleges 2026 adds a rule that runs the other way: when marking NEA, schools “must not use AI as the sole marker.”
If you are moving on to university or applying now
None of this carries over to higher education. Universities are not JCQ centres; each sets its own misconduct rules under a different set of bodies, which who regulates AI detection in UK universities explains. Sixth-formers writing an application this autumn face a third system again: what UCAS does with your personal statement covers its similarity checks and declaration.
Where we stand
HumanFlow’s detector and humanizer read English only, which rules them out for any NEA submitted in Welsh. Neither appears in JCQ’s list, and we would not claim JCQ’s authority for them.
A rewrite does not change your declaration. JCQ lists paraphrasing AI-generated content as misuse in its own right, so text a model produced is not made your own work by being reworded, by us or by anyone. If you used AI, acknowledge it in the format above before you sign. If you are a teacher, JCQ’s own caveats are the right frame for any score, ours included: one input among several, weaker on edited text, and never a substitute for knowing the student.
This is not legal advice or advice from an awarding body. Your exams officer and your subject teacher are the people who can say how your school applies these rules.